The effect of pH, grain size, and organic ligands on biotite weathering rates

Biotite dissolution rates were determined at 25 C, at pH 2–6, and as a function of mineral composition, grain size, and aqueous organic ligand concentration. Rates were measured using both openand closed-system reactors in fluids of constant ionic strength. Element release was non-stoichiometric and followed the general trend of Fe, Mg > Al > Si. Biotite surface area normalised dissolution rates (ri) in the acidic range, generated from Si release, are consistent with the empirical rate law:


INTRODUCTION
The weathering of silicate minerals controls the chemical composition of natural waters, supplies nutrients to the biosphere and, over geological time-scales, weathering is a crucial regulatory process of the global carbon cycle (Berner and Berner, 1996). Among silicates, the weathering of phyllosilicates has a particularly strong influence on geochemical cycles due to their abundance, large specific surface areas, and high ion-exchange capacities (Sposito, 1984;Davis and Kent, 1990;Drever, 1997;Bowser and Jones, 2002). Among phyllosilicates, biotite is the key source of primary (K) and secondary (Fe, Mg, Al) nutrients in soils and freshly exposed rocks, particularly in Boreal regions. Overall, it has been estimated that biotite makes up $7% of the exposed crustal surface (Nesbitt and Young, 1984). During the past few decades, many field and laboratory studies have shown that the biosphere is crucial in controlling mineral weathering and therefore influencing geochemical processes (Beerling and Berner, 2005;Taylor et al., 2009), yet often a validation of the magnitude of these processes is lacking.
For biotite, a number of experimental studies quantifying its abiotic dissolution have revealed that at a given pH, the dissolution rates can vary by up to 1.5 orders of magnitude (Acker and Bricker, 1992;Turpault and Trotignon, 1994;Kalinowski and Schweda, 1996;Malmströ m et al., 1996;Malmströ m and Banwart, 1997;He et al., 2005). From studying other mineral systems, these variations may be explained by several possible factors; (a) differences and variations in bulk chemical composition, (b) differences in particle size and geometry, (c) differences in experimental approaches or durations, and (d) variations in chemical compositions and concentrations of the experimental fluids Brantley, 2003;Kö hler et al., 2005;Schott et al., 2009;Zhang and Lü ttge, 2009;Arvidson and Lü ttge, 2010;Fischer et al., 2014;Rimstidt, 2014). The purpose of this study is to document how each of these factors affects biotite dissolution.
Biotite is a subgroup of the mica mineral group, which forms a solid solution series with iron and magnesium end members annite (K Fe 3 AlSi 3 O 10 (OH) 2 ) and phlogopite (K Mg 3 AlSi 3 O 10 (OH) 2 ). The variable concentrations of Fe and Mg in natural biotite samples, in conjunction with the substitution of a wide variety of other elements in the octahedral and tetrahedral silicate sheets, may lead to differences in element release rates when normalised to the biotite formula. Therefore, to test the effect of mineral composition on dissolution rates, we present here data from the dissolution of two chemically distinct biotite samples.
Moreover, biotite is highly anisotropic in terms of surface reactivity, with edge (h k 0) surfaces being between 30 and 300 times more reactive than basal (0 0 1) surfaces (Turpault and Trotignon, 1994;Hodson, 2006). This implies that grain shape and size play a crucial role in measuring and interpreting dissolution rate data. Surprisingly however, the majority of published biotite dissolution rates are normalised to the total surface area, and the effect of geometric anisotropy on reactivity has not been addressed experimentally. Therefore, in this study, we present data from experiments performed with three different biotite size fractions, each with distinct geometric (edge to basal surface) ratios to better understand the control of grain size and geometry on biotite dissolution kinetics.
Existing biotite dissolution kinetic studies have employed a variety of experimental approaches. These include (a) open-system experiments performed in fluidized-bed reactors and flow-through columns (Acker and Bricker, 1992), dialysis-cell reactors (Kalinowski and Schweda, 1996), thin-film flow-through reactors (Malmströ m et al., 1996;Malmströ m and Banwart, 1997), and continuous-flow reactors (Hodson, 2006;Voinot et al., 2013), and (b) closed-system experiments where biotite was reacted in batch mode with a variety of solutions (Turpault and Trotignon, 1994;Balland et al., 2010). All of the open-system experiments were performed for days to months, allowing the determination of steady-state, far from equilibrium dissolution rates. However, significant changes in both surface area and reacting mineral composition can occur during such open-system experiments, requiring the careful consideration of surface area normalisation (Brantley, 2003;Hodson, 2006). Often, rate data from open-system experiments are normalised to an initial surface area and initial mineral composition, and changes in either of these parameters are not considered. Conversely, closed-system experiments are usually performed over shorter timescales (hours to days) and these allow primarily the characterisation of the initial stages of dissolution. However, rates determined from these shorter, closed-system experiments can be significantly affected by rapidly changing fluid compositions, and rates can be influenced by secondary phase precipitation Brantley, 2003). To overcome this lacuna, we have performed both mixed-flow (open-system) and batch (closed-system) experiments with the same biotite, at equivalent conditions and derived, compared, and contrasted dissolution rates from both systems.
Finally, most past studies only focussed on the abiotic biotite dissolution. In natural settings, however, biotite is a prime source of nutrients (K in particular) and thus biotic processes may play an important role in element release. Several recent studies have demonstrated that microorganisms influence phyllosilicate weathering through a combination of physical and chemical alteration, or 'bio-fracking' (Balogh-Brunstad et al., 2008;Bonneville et al., 2009Bonneville et al., , 2011Hopf et al., 2009;Balland et al., 2010;Saccone et al., 2012;Gazzè et al., 2012Gazzè et al., , 2013. One bio-mediated mechanism is changing the chemical composition of the local environment, i.e., localised acidification (Balogh-Brunstad et al., 2008;Bonneville et al., 2011). Such acidification is the result of microbial/fungal respiration and the release of organic acids in the near environments of living microorganisms (Bonneville et al., 2011;Gazzè et al., 2013). The effects of which have been considered in several studies attempting to quantify the effect of organic ligands (low molecular weight organic acids and siderophores) on biotite dissolution (e.g., Drever and Stillings, 1997;Balland et al., 2010;Voinot et al., 2013). The results suggest a catalysing effect of organic ligands on phyllosilicate weathering (i.e., a doubling of the dissolution rates) in the presence of high concentrations of organic ligands (i.e., >10 À3 M, Balland et al., 2010). However, the effect of organic ligands on mineral dissolution at the low concentration levels typically found in soils (10 À5 to 10 À6 M, Jones, 1998;Ullman and Welch, 2002), and known to be secreted by microorganisms (<10 À3 M, Adeyemi and Gadd, 2005) is unclear. Furthermore, only a few data sets testing the effects of organic ligands on biotite dissolution exist (Balland et al., 2010;Voinot et al., 2013). Balland et al. (2010) describe the release of Fe from biotite and phlogopite in closed-system experiments from approximately pH 2-7, both abiotically and in the presence of relatively high concentrations (10 À2 and 10 À3 M) of gluconic acid and citric acid. However, data on the release of other elements is absent. Voinot et al. (2013) present a dissolution rate enhancement of almost an order of magnitude in the presence of citric acid at pH 3 and 4.5, compared to open-system abiotic biotite dissolution experiments. Here, we have reacted biotite with three organic ligands (citric acid, oxalic acid, and the siderophore desferrioxamine B, DFOB) and assessed their influence on the release of elements (Al, Fe, Mg and Si) and biotite dissolution rates between pH 2 and 6.
In summary, the objectives of this study were to quantify the control of (a) bulk mineral chemistry, (b) grain size and geometry, (c) experimental approach, and (d) aqueous fluid composition, including organic ligands, on biotite dissolution rates. Experimental results were used to illustrate the effect of the presence of microorganisms both in laboratory and field settings, aiding our understanding of natural soil based processes.

Mineral samples
The chemical composition and stoichiometric formulas of the two biotite samples (Grasåsen and Moen, both Norway) used in this study have previously been described (Moen - Bonneville et al., 2009;Grasåsen -Bray et al., 2014) and are summarised in Table S1. Each sample was crushed separately using a jaw crusher, pestle and mortar, ball mill, and agate disk mill, removing inclusions throughout. The crushed biotite samples were subsequently sieved to obtain the 75-150 lm (Moen), and 25-53, 53-180, and 180-500 lm (Grasåsen) size fractions. Each size fraction was then washed ultrasonically 10 times in 18.2 MO water, separated by gravitational settling, and dried at 40°C for 48 h.
Post washing, an aliquot of each size fraction was imaged using a field emission gun scanning electron microscope (FEG-SEM, FEI Quanta 650 equipped with an Oxford X-Max silicon drift detector, SDD, operated at 20 kV) on a stage set at a 45°angle to determine grain geometries. From these measurements the geometric specific surface area (SSA geo , m À2 g À1 ), and the average edge (SA edge %) and basal plane (SA basal %) proportions of each size fraction were determined (Supplementary Information). The specific surface area (SSA BET , m À2 g À1 ) of each size fraction, and of the post reaction solids, was measured via a 9 point N 2 adsorption isotherm between 0.05 and 0.25 p/p o (equilibrium pressure/saturation pressure) at 77 K using a Micromeritics Gemini V analyser. Samples were purged with N 2 for 20 h before analysis and the surface area was calculated using the BET method (Brunauer et al., 1938) based on an assumed cross section of adsorbed N 2 of 0.162 nm 2 . Replicate measurements of 76 sub-samples of the 53-180 lm size fraction (measured at the start of each closed-system Grasåsen biotite experiment) yielded a relative standard deviation (RSD) of 4.6%. The SSA BET , SSA geo , SA edge , and SA basal of each size fraction are given in Table 1, with the details of calculation of grain geometries presented in the Supplementary Information. Images of the post washing, pre-reaction biotite grains are shown in Fig. S1.

Closed-system experiments
In total, 91 closed-system experiments were conducted in 1 L high density polyethylene bottles (HDPE, Nalgene), with each experimental condition for the Grasåsen biotite performed in triplicate. The bottles were placed in a 25 ± 1°C incubator for up to 12 h where they were constantly shaken at 140 rpm, each bottle was also vigorously shaken manually at least once every hour. For each experiment, 1-2 g of biotite was added to 1 L of experimental fluid. Two types of experimental fluids were prepared: (i) one set of experiments (using the Moen biotite) was performed with aqueous 0.003 moles kg À1 CaCl 2 fluids (ionic strength = 0.01 M, AnalaR NORMAPUR, VWR), with the pH adjusted to between 2 and 6 using HCl/NaOH (Aristar/AnalaR NORMAPUR, VWR); (ii) a second, larger set of experiments (using the Grasåsen biotite) was performed with fluids that were buffered at a selected pH from 2 À11.8% k a n = 3; b n = 76; c n = 3; d n = 1; e n = 50; f n = 50; g n = 25; h n = 18; i Post reaction samples from ligand free closed-system experiments at pH $2.5 (CS-G01-03); j Post reaction samples from ligand free closed-system experiments at pH $6 (CS-G19-21); k Post reaction samples from DFOB closed-system experiments at pH $6 (CS-G49-51).
to 6 with potassium hydrogen phthalate (KHP, C 8 H 5 KO 4 , CertiFied AR, Fisher). KHP was used as a background electrolyte to fix the ionic strength and to buffer pH, because initial experiments with CaCl 2 showed a strong pH buffering capacity of the Grasåsen biotite due to a CaCO 3 impurity of approximately 3.7 wt.%, as previously described (Bray et al., 2014). Experiments were also conducted using both types of experimental fluids to which 500 lmoles kg À1 of either citric acid (C 6 H 8 O 7 ), DFOB (C 25 H 48 N 6 O 8 ), or oxalic acid (C 2 H 2 O 4 ) were added. Details for all experimental conditions are given in Table 2. The temporal pH evolution in each experimental condition (with each biotite and experimental fluid matrix, with or without organic acids added) was monitored in additional experiments that were not used for dissolution measurements. In these experiments a semi-micro, epoxy pH electrode (VWR 662-1767), calibrated using NIST traceable buffer solutions of pH 4.01, 6.99, and 9.99 (Fisher Scientific UK) at 25°C, and connected to an Orion Dual Star pH meter (Thermo Scientific), was constantly immersed in the experimental fluids and the pH was logged every 5 min during the 12 h experiment. Furthermore, in all dissolution experiments the initial, intermediate, and final pH was measured. The maximum pH variance in each experiment was ±0.08 pH units (based on average of 2r of pH for each experiment, 95% confidence). During each 12 h closed-system experiment, and with decreasing frequency as each experiment progressed, up to 20 sample aliquots containing 5 mL of a mixture of experimental fluid and biotite particles were removed using a 5 mL syringe from the 1 L containers. Removed samples were immediately filtered through 0.2 lm polyethersulfone filters (Sartorius) and quenched in 3% HNO 3 (Aristar, VWR), HCl (Aristar, VWR) or NaOH (AnalaR NORMAPUR, VWR) depending on subsequent analysis. At the end of each 12 h experiment, the remaining fluids were separated from the reacted biotite by centrifugation. All fluids were analysed as described below, and the reacted biotite powders were dried for subsequent imaging and analysis. Fluid samples were analysed for Al, Fe, Mg, and Si by inductively coupled plasma mass spectrometry (ICP-MS, Agilent 7500ce with integrated auto-sampler, analytical error 6.3%). The detection limit for Al, Fe, Mg, and Si, determined from calibration curves, was 0.8, 0.6, 5.3, and 4.7 lg kg À1 , respectively. The average analytical uncertainty, measured against two certified reference materials for Al, Fe, and Mg (SLRS-5 CRM, National Research Council Canada; NWTM-27.3, LGC Standards), and an in-house standard for Si, was determined to be 5.7%, 6.7%, 1.1%, and 7.9%, respectively. To cross-check our data quality, Mg was also analysed by atomic absorption spectroscopy (AAS, Analytik Jena Contraa 700, maximum analytical uncertainty of 2.8%). The detection limit for Mg by AAS was 2.6 lg kg À1 with a maximum analytical uncertainty of 7.8%.

Open-system experiments
Both the Moen and Grasåsen biotite samples were reacted in open-system experiments with aqueous fluids of pH 2-4 and at constant ionic strengths. Two distinct mixed-flow reactor systems were used.
The Moen biotite was reacted with an aqueous CaCl 2 solution of 0.01 M ionic strength at pH 3.3 in a 300 mL titanium Parre mixed-flow reactor at controlled temperature, pressure, and under constant stirring and continuous pH monitoring following the approach described in Wolff-Boenisch et al. (2004). Briefly, approximately 6 g of the 75-150 lm biotite size fraction was reacted in a stirred (400 rpm) 300 mL mixed-flow reactor, for up to 260 h with a fluid continuously pumped through the reactor with a high performance liquid chromatography (HPLC) pump at flow rates from 1.2 to 2.2 g/min. At the system outlet, the fluid passed through a 2 lm titanium filter to prevent the escape of solids. The outflow fluids were further filtered using 0.2 lm polyvinylidene fluoride (PDVF) Whatmane filters and collected in acid washed HDPE sample bottles, pre-acidified with HNO 3 (Aristar, VWR). The solids were recovered at the end of each experiment by centrifugation and dried for further analyses.
The Grasåsen biotite was reacted with fluids at pH from 2 to 4 and at an ionic strength of 0.01 M, fixed using HCl, NaCl, or KHP, with the KHP used as a pH buffer and ionic strength matrix (see above). Experiments were conducted in a water bath at a constant temperature of 25 ± 1°C in 35 mL Teflon mixed-flow reactors that were continuously stirred with floating Teflon magnetic stirring bars. Each reactor consisted of identical base and head compartments (Kö hler et al., 2005;Chaïrat et al., 2007;Flaathen et al., 2010;Declercq et al., 2013) and the added biotite powder (1-1.5 g) was kept in the reactor using 0.45 lm GN-6 Metricel MCE membrane filters. During the experiments (up to approximately 800 h) fluids were pumped through the reactors via a peristaltic pump at rates from 0.1 to 0.6 g/min. Samples were collected in HDPE bottles, pre acidified with 3% analytical grade HNO 3 . Fluid pH was measured at the start of each experiment and in each fluid sample taken during the experiments, using a Metrohm 713 pH Meter, calibrated with certified reference buffer solutions of pH 4.005, 6.866, and 9.183 (CertiPUR, Merck) at 25°C. pH measurement precision was 0.02 pH units. Dissolved silica concentrations in the outlet fluids were determined using the colorimetric molybdate blue method (Bran & Luebbe AutoAnalyzer 3, maximum analytical uncertainty ±3.5%). Concentrations of Fe, Mg, and Al were determined by AAS using a Perkin Elmer 5100 PC spectrometer with maximum analytical uncertainty of 2.8%. The details of the open-system experiments are listed in Table 3.

Experimental fluid modelling
The speciation of organic ligands and dissolved elements in the experimental fluids and the saturation state of the fluids with respect to potential secondary phases were calculated using the PHREEQC3 computer code and the minteq.v4 database (Parkhurst and Appelo, 2013). The results of these calculations were used to interpret the experimental results as discussed below.

Closed-system experiments
The measured aqueous element concentrations (c i,t ) released as a function of time in all closed-system experiments are provided in Table S2. To quantify the rates of biotite dissolution the number of moles of each element, i, in the fluid phase at time t (m i,t ), normalised to biotite stoichiometry (Table S1) and the surface area present in the reactor (from the initial BET specific surface area, s) (Table 1), is defined as m 0 i,t , and calculated through: where c i,t (moles kg À1 ) designates the measured fluid concentration of element i at time t, V t refers to the mass (kg) of aqueous fluid in the reactor at time t, s represents the initial specific surface area of biotite powder (m 2 g À1 , SSA BET ), M the mass of biotite powder (g), and n i the stoichiometry of the ith element in the biotite formula. Resulting values of m 0 i;t are shown in Figs. 1 and 2. Element release during the closed-system dissolution experiments was highly pH dependent and, in part, non-stoichiometric. In organic ligand-free experiments at pH 2.5, when normalised to the dissolving biotite stoichiometry, the release of Mg, Fe, and Al was largely stoichiometric with respect to one another, while Si release was 2.5-3 times lower (Fig. 1A). With increasing pH the released element concentrations decrease ( Fig. 1B) with the final concentrations (after 12 h) at pH $6 being $1 order of magnitude lower than those at pH 2.5 (Fig. 1C). Increased pH also results in more pronounced non-stoichiometric release, with preferential Mg release over other elements at pH > 4 (Fig. 1C).
In experiments at approximately pH 4, where the same Grasåsen biotite but different particle size fractions were used (Tables 1 and 2), the element release profiles, plotted as biotite stoichiometry and surface area normalised values, show a marked increase in element concentration with decreasing particle size (Fig. 1D, Fig. S2). For aluminium, magnesium and silica in all cases, after 12 h, the smallest biotite size fraction (25-53 lm) released Al, Mg and Si between 0.5 and 2 times faster than the 53-180 and 180-500 lm size fractions respectively ( Fig. 1D and Fig. S2A, C and D). In contrast, iron release ( Fig. S2B) appeared independent of particle size. One possible explanation for this is the precipitation of the dissolved iron in secondary phases. The fluids at the end of the organic ligand-free experiments (12 h) for each size fraction at pH 4 (experiments CS-G07 to G15, Table 2) were calculated to be supersaturated with respect to Fe-phases (ferrihydrite, goethite and hematite; Table S3). However, FEG-SEM imaging of the reacted biotite solids, even at the highest resolution ($20 to 50 nm particles), did not reveal any secondary phases. Such phases were only observed on particles retrieved from the organic ligand-free closed-system experiments at pH 6 (see CS-G19-21, Table 2, Fig. S1D) where the solubility of secondary Fe and Al phases is minimal. Precipitation of such secondary phases could result in Table 3 Details of the  Table 2), only revealed specific surface area changes of +0.6% and À2.3%, respectively, which are both within the uncertainty of measurement (9.2%, two relative standard deviations based on 76 replicates of the original material, Table 1).
In experiments with organic ligands present, at low pH (pH 2.5, Fig. 2A), there was no appreciable effect on Si release. However, at pH 3 and above, the presence of organic ligands increases element release relative to ligand-free experiments. This is most evident for Fe and Al ( Fig. 2B and C) where, compared to the ligand-free system, the end-experiment Al and Fe concentrations in the presence of all organic acids were between 2 and 6 times higher. A similar pH dependent behaviour is evident for the final Mg concentration (t = 12 h) shown in Fig. 2D. These element release profiles are representative of the overall trend of the effect of ligand presence observed for Al, Fe, Mg, and Si at pH values between pH 2.5 and 6. All other element release profiles are presented in Fig. S3.
Evaluating dissolution rates from closed-system experiments is challenging because kinetic calculations must consider the constantly changing fluid composition and the effect of possible saturation controlled back reactions Brantley, 2003). The release of elements into the fluid during our short-term, 12 h, closed-system experiments followed a trend typical for most closed-system dissolution studies (e.g.,  where rapid release of elements occurs in the first stage of the reaction followed by a slower, decreasing release with time. The element release profiles can be modelled (Fig. 3A) according to: where k 0 designates a rate constant (moles m À2 s Àq ), t refers to time in seconds, and q stands for the exponential decay of the release rate (dimensionless). In Eq.
(2), the values of k 0 and q can be derived through a linear regression of the logarithm of m 0 i,t and t, where q corresponds to the slope of the regression and log k 0 , its intercept (log t = 0, t = 1 s). The values of k 0 and q for each experimental condition are listed in Table 2.
The element release profile shown in Fig. 3A is typical for the closed-system dissolution experiments. Such profiles can be interpreted to stem from (1) the initial rapid release of elements from fine particles or highly reactive sites, and (2) the slowing of rates due to an approach to equilibrium, or (3) a slowing of rates due to the increase of one or more aqueous species that inhibit dissolution Gautelier et al., 2007;Saldi et al., 2007Saldi et al., , 2010Oelkers et al., 2008). Based on these profiles, we determined rates directly from the derivative of the element masses versus time throughout each experiment. The dissolution rate can be calculated analytically from the derivative of Eq. (2): where r i,t represents the dissolution rate (moles biotite m À2 s À1 ) with respect to the ith element at time t. The calculated biotite dissolution rates for each closed-system experiment at t = 1 s and t = 12 h are shown in Table 2. The results reveal that for all elements the dissolution rates decrease with time, with the rate at 12 h approximately 4 orders of magnitude slower than the initial dissolution rate (t = 1 s). The causes of the differences in these rates will be discussed below.

Open-system experiments
In the organic ligand and ligand-free open-system experiments the element release rates approached a near steady-state after at least 100 h (Fig. 3B). Steady-state was defined by the outlet fluids maintaining a constant concentration of Al, Fe, Mg, and Si, within 5% (analytical and experimental uncertainty) for a minimum of 3 residence times (reactor volume/flow rate). In the case of the Grasåsen biotite at pH 2, steady-state was attained after $125 h. Based on this, the average element concentrations released after 125 h (Table 3) were used to calculate steady-state biotite dissolution rates according to: where r i denotes the surface area normalised steady-state dissolution rate (moles biotite m À2 s À1 ), f symbolises the fluid flow rate (kg s À1 ), c i signifies the aqueous concentration of the ith element in the outlet fluid (moles kg À1 ), s stands for the initial specific surface area (m 2 g À1 , SSA BET ) of the mineral powder of mass M (g), and n i the stoichiometric coefficient of the ith element in the biotite formula (Table S1). The calculated biotite dissolution rates for each open-system experiment are listed in Table 3.

pH and organic ligand dependent element release
Consistent with the general trend of multi-oxide silicates Schott et al., 2012;Stockmann et al., 2013;Rimstidt, 2014), the Grasåsen and Moen biotite dissolution rates decrease with increasing pH from pH 2-6. The calculated open and closed-system organic ligand-free dissolution rates are plotted as a function of pH in Fig. 4A, along with biotite dissolution rates from the literature. The shaded area in Fig. 4A is used in Fig. 4B-D to highlight the range of kinetic data presented in the literature as a function of pH (Acker and Bricker, 1992;Kalinowski and Schweda, 1996;Malmströ m et al., 1996;Malmströ m and Banwart, 1997;Balland et al., 2010;Voinot et al., 2013). The closed-system rates plotted are those calculated for t = 12 h, as these are most consistent with open-system and literature data. It is worth noting that, at any given pH, biotite dissolution rates from the literature can range over 1.5 orders of magnitude.
This range of reported rates can be attributed to a number of differences. Firstly, differences in the bulk chemical composition of each biotite sample may result in a range of dissolution rates. The rate data from the literature presented in Fig. 4A span biotite samples from 4 locations: Bancroft, Ontario, Canada (Acker and Bricker, 1992;Voinot et al., 2013); Moen, Norway (Kalinowski and Schweda, 1996;Malmströ m et al., 1996;Malmströ m and Banwart, 1997; current study); Raze, Limousin, France (Balland et al., 2010) and Grasåsen, Norway (this study). However, even when a biotite sample was used from the same location some compositional differences remain. For example, in the case of the biotite from Bancroft, Ontario, the reported Mg and Fe present in each sample vary. Acker and Bricker (1992) report moles of Mg and total Fe in one mole of biotite as 1.59 and 1.12 respectively, while in the Bancroft, Ontario biotite used by Voinot et al. (2013) Mg and Fe in the formula are reported as 2.06 and 0.73 moles respectively. Secondly, the anisotropic nature of biotite has previously been described (Turpault and Trotignon, 1994;Hodson, 2006), with differences in grain size, surface area and geometries of the reacting biotite all likely influencing the measured dissolution rates. Thirdly, different experimental approaches, including experiment duration, may also influence dissolution kinetics. For example, for illite, Kö hler et al. (2005) reported decreasing dissolution rates with time during flow-through experiments, and attributed these to the changes in clay morphology and a decrease in the number of reactive edge sites. For biotite, the duration of flow-through biotite dissolution experiments in the literature (Fig. 4A) ranges from $300 (Acker and Bricker, 1992) to 3000 h (Kalinowski and Schweda, 1996). Finally, variations in the chemical composition of the experimental fluids can influence dissolution rates, particularly in closed-system experiments where fluids may become supersaturated with respect to secondary phases.
To evaluate how these various parameters have affected the dissolution rates of the Grasåsen and Moen biotite in the current study, the dissolution rates are described as a function of pH using an empirical rate law for acid promoted dissolution (e.g., Lasaga, 1998): where r i again refers to the specific dissolution rate of biotite based on the ith element, k H,i , corresponds to a rate constant, and x refers to a reaction order with respect to proton activity, a H þ .
Using Eq. (5), the rate constants and reaction orders derived from the ligand-free experiments for both the Grasåsen and Moen biotite samples were determined by linear-least-square regression of log r i against pH (Table 4; Fig. 4A). Results are comparable with the rate constants and reaction orders previously reported for biotite (Acker and Bricker, 1992;Kalinowski and Schweda, 1996;Malmströ m and Banwart, 1997).
The data derived from the biotite dissolution experiments performed in the presence of citric acid, oxalic acid, and DFOB (Fig. 4B-D) generally lie within the shaded region of the ligand-free biotite dissolution rates shown in Fig. 4A. At low pH (<4), biotite dissolution rates in the presence of citric acid, oxalic acid, and DFOB were within the range of the organic ligand-free data from this study, and the literature. Above pH 4 a slight enhancement in dissolution rate in the presence of all organic ligands could be observed compared to the ligand-free data. However, the ligand dissolution rates are within the range of abiotic biotite dissolution rates reported previously, and in this study ( Fig. 4B-D). In comparison, Balland et al. (2010) conducted biotite dissolution experiments in the presence of citric and gluconic acid at ligand concentrations between 10 À3 and 10 À2 M. Compared to the data in the current study, Balland et al. (2010) reported an enhanced ligand promoted release rate for Fe, which was up to 2 orders of magnitude faster than their corresponding ligand-free rates. Similarly, Voinot et al. (2013) reported an order of magnitude enhancement in biotite dissolution rates in the presence of citric acid, but they saw no enhancement in the presence of the siderophore DFAM-B.
In the current study, the minor increase of element release from biotite in the presence of organic ligands may be explained by: (1) the relatively low, but more realistic, concentrations of organic ligands used in the experimental fluids, and (2) that biotite dissolution rates have been computed from Si release rather than the release of other aqueous metals.
The effect of organic ligand concentrations on rates is not easy to compare. We used concentrations of citric and oxalic acid of 500 lM each, while DFOB was 500 lM and 145 lM in the closed-and open-system experiments, respectively. These values were chosen to be consistent with organic ligand production values reported for soil-dwelling microorganisms in the literature (Lapeyrie et al., 1987;Lapeyrie, 1988;Adeyemi and Gadd, 2005). However, comparing our data with previous biotite studies where organic ligands were reported to enhance dissolution, revealed that the concentration of citric acid used in the current study was between 2 and 20 times lower than the concentration used by Balland et al. (2010), while the citric acid concentration used by Voinot et al. (2013) was not reported.  showed that a system requires a minimum concentration of organic ligands (P10 À3 M) that is between 2 and 7 times higher than the value used here, before an appreciable ligand promoting dissolution effect (i.e., a doubling of dissolution rate) is measurable. Such a critical ligand concentration was also observed in dissolution studies of basaltic glass, where Oelkers and Gislason (2001) found a 2-fold increase in basaltic glass dissolution rates by adding 1000 lM oxalic acid to a pH 3 solution, whereas Wolff-Boenisch et al. (2011) did not find any change in dissolution rates of the same basaltic glass in the presence of 120 lM oxalic acid, at similar pH. The fact that neither Voinot et al. (2013), nor this study, finds any DFOB effect on the Si release rates may well be explained by the similarly low concentrations used in both studies.
Differences in rates between our study and literature studies (Fig. 4) may also stem from the fact that we calculated rates based on the temporal evolution of Si concentrations in the experimental fluid. One of the reasons for the reported effect of citric acid on biotite dissolution rates in the study of Balland et al. (2010) is likely due to their use of the dissolved Fe concentrations, a metal to which citrate has a strong affinity, for their rate calculations. This is because in the absence of organic ligands at slightly acidic pH the solubility of iron is low and enhanced by low concentrations of citric acid. This effect likely leads to the apparent faster rate reported by Balland et al. (2010). This conclusion is strengthened through speciation and metal complexation calculations (Fig. S4) for (A) phthalic acid, (B) citric acid, (C) oxalic acid and (D) DFOB as a function of pH. These plots were calculated from the acid dissociation and stability constants of each ligand (Table S4), the concentrations of Al, Fe and Mg at the end of the pH 2.5 closed system experiments (12 h, Table S2) and using the PHREEQC 3 computer code (Parkhurst and Appelo, 2013). Above pH 2-3 in the presence of citric acid (Fig. S4B), the majority of Fe and Al in the fluid forms citrate complexes. Conversely, Mg-citrate complexes do not become dominant until above pH 6. This poor affinity for the formation of Mg-organic ligand complexes is also reflected in the oxalic acid and DFOB speciation. At pH 1, Fe-oxalate and Fe-DFOB complexes are the dominant form of dissolved Fe, while Al-oxalate, Al-DFOB complexes dominate above pH 2.5 and 2 respectively ( Fig.S4C and S4D). We used the same code to calculate the saturation state of the experimental fluids with respect to potential secondary phases in all experiments (Table S3). Compared to a fluid calculated using NaCl as a background electrolyte, the presence of phthalate (from KHP) has little effect on Al and Fe secondary phase saturation. This is due to the poor affinity of phthalate for metals. Table 4 Apparent rate constants k H (10 À9 moles m À2 s À1 ), and reaction orders x, with respect to hydrogen ion activity from the Moen and Grasåsen element release data. SE represents the standard error of the apparent rate constants and reaction orders. In the organic ligand-free experiment, experimental fluids across the pH range studied are supersaturated with respect to iron phases (ferrihydrite, goethite, and hematite), and supersaturated with respect to the Al phase gibbsite at pH > 5. However, the calculated saturation indices for Fe and Al phases are reduced in the experimental fluids with citric acid, oxalic acid and DFOB (Table S3). At least one Fe phase is supersaturated at all pH in the presence of citric acid, and above pH 5 in the presence of oxalic acid. However, Fe phases are undersaturated at all pH in the presence of DFOB, highlighting the powerful chelating ability of the siderophore. Apart from at approximately pH 6 in the presence of oxalic acid, Al phases are all undersaturated in the presence of the organic ligands. The calculated positive saturation values in some near neutral experiments, with respect to secondary iron and aluminium phases, could be due to the presence of nanoparticulate or colloidal iron and aluminium. Particles less than 200 nm in size could have passed through the 0.2 lm filters during sampling, and subsequently been measured as solution based Fe and Al during analysis. However, as the point of zero surface charge for biotite is pH 3 (Bray et al., 2014), and pH 7-9 for Fe and Al oxides and oxyhydroxides (Kosmulski, 2009), the Fe and Al nanoparticles are likely to adsorb to the biotite surface between pH 3 and 7 due to the opposing surface charges. Therefore, the positive saturation indices likely indicate that Fe and Al phases may have precipitated during some of the closed system experiments, however these were only visible by scanning electron microscopy at pH 6 in the ligand-free experiments (Fig. S1D, CS-G19-21), and did not result in significant specific surface area changes in the post reaction samples (Table 1). These calculations highlight that the presence of organic ligands decreases secondary phase saturation, allowing the further release of Al and Fe into the fluid phase at a given pH, compared to the organic ligand-free experimental fluids. The mechanism by which an organic ligand influences mineral dissolution has been described as complexation, either direct surface complexation or indirect solution based chelation and the associated effect on the solution affinity and/or activity of dissolution inhibitors, such as Al, (c.f. Drever and Stillings, 1997;. In the case of biotite, it is likely that metal-ligand complexation occurs mainly in the fluid phase. In our previous study on the Grasåsen biotite (Bray et al., 2014) we demonstrated that the isoelectric point of biotite is at pH 3. Below this pH, the biotite surfaces have a net positive charge, above this pH, a net negative charge. Above pH 3 the dominant species of citric and oxalic acid are negatively charged such that interactions with the negatively charged biotite surface are likely limited. However, the dominant DFOB species in the range of pH of this study (2-6) is the positively charged H 4 DFOB + species that might have been able to adsorb to biotite. However, although it has been demonstrated previously that the presence of DFOB promotes Al and Fe release to solution (Hersman et al., 1995;Holmén and Casey, 1996;Kraemer et al., 1999;Cocozza et al., 2002;Cheah et al., 2003;Kraemer, 2004;Buss et al., 2007;Reichard et al., 2007;;Wolff-Boenisch and Traina, 2007a,b), its actual adsorption to the dissolving matrix has been discussed to be insufficient (3%) to account for the observed increase in dissolution rates, probably because of steric hindrance (Wolff-Boenisch and Traina, 2006).

Surface area and dissolution
The surface area normalised dissolution rates of the three Grasåsen biotite size fractions do not have a linear relationship with the total mineral surface area present in each experiment (SA BET m 2 , Fig. S5). This observation may stem from the anisotropic nature of the biotite surface reactivity (e.g., Turpault and Trotignon, 1994;Hodson, 2006). The logarithm of the dissolution rates appear to exhibit a linear correlation with the fraction of geometric specific surface attributed to the biotite edges (h k 0, SA edge %, Fig. S5, Table 1), demonstrating the importance of grain geometry on the overall dissolution rate. Like biotite, other sheet silicates have been shown to primarily dissolve parallel to the basal surface, i.e., at the (h k 0) surfaces (Kaviratna and Pinnavaia, 1994;Turpault and Trotignon, 1994;Bosbach et al., 2000;Bickmore et al., 2001Bickmore et al., , 2003Aldushin et al., 2006;Hodson, 2006;Saldi et al., 2007;Cappelli et al., 2013).
From the grain geometry measurements made using FEG-SEM ( Fig. S1 and Table 1), the overall biotite dissolution rate constant at a single pH value (k 0 Si , Eq. (2)) can be considered to be the sum of two contributions, in accord with (e.g., Hodson, 2006): where k 0 Si;edge and k 0 Si;basal (moles m À2 s Àq ) correspond to the rate constants of the dissolution of edge and basal surfaces respectively, and SA edge and SA basal designate the fraction of edge and basal surfaces of the total geometric surface area (Table 1). The values of k 0 Si;edge and k 0 Si;basal were obtained from extrapolating the linear regressions of k 0 Si with SA edge % and SA basal % to 0 % SA basal and SA edge , respectively. At pH $4 the value of k 0 Si;edge is 36.7 Â 10 À7 (moles m À2 s Àq ), approximately 120 times greater than k 0 Si;basal , 0.31 Â 10 À7 (moles m À2 s Àq , Table 5). The relative reactivities of the edge and basal surfaces calculated for the other elements also differ for edge or basal contributions. For example, the values of k 0 Al;edge and k 0 Mg;edge are 102 and 45 times greater than k 0 Al;basal and k 0 Mg;basal , respectively. The relative reactivity of the edge and basal surfaces could not be calculated for Fe, as iron release was independent of grain geometry (Fig. S2B). These calculations, although limited due to the need to extrapolate experimental data, suggest that the relative reactivities are element specific, as observed previously (Turpault and Trotignon, 1994;Malmströ m and Banwart, 1997). Nevertheless, the calculated relative reactivity ratios of (h k 0):(0 0 1) for each element (excluding Fe) fall within the range given by Trotignon (1994) of 30-300, andHodson (2006) of 71-132. However, the range presented in this study is pH specific and considering that the studied biotite changes its surface properties (surface charge and composition) with pH (Bray et al., 2014), the surface reactivities, both for basal and edge planes, will also invariably be affected.
The total concentration of a given element released from biotite into the fluid phase (m 0 i,t ) is controlled by the portion of edge and basal surface in each particle size fraction, and is the sum of these two distinct contributions. The contributions of the edge and basal surfaces to the total released Si at pH $4 is shown for each size fraction in Fig. 5, calculated as in Eq. (7): The symbols in Fig. 5 are the experimentally measured, surface area and stoichiometry normalised concentrations of Si for each size fraction: (A) 25-53, (B) 53-180, and (C) 180-500 lm (experiments CS-G07-15). The solid black line in Fig. 5A, B, and C is the modelled fit of the experimental data (Eq. 2), which is the sum of the modelled contribution from the edge (long dash) and basal (short dash) surfaces.
When the modelled summed m 0 Si values are plotted against the experimentally measured m 0 Si values a good fit is achieved (Fig. 5D). It is worth noting that the values of k 0 edge and k 0 basal were calculated using a fixed average value of q (Eq. 1) as this parameter varies by 0.06 as a function of grain size for silicon (Table 2). However, the modelled concentrations (Eq. 7, Fig. 5) were calculated using the values for q provided in Table 2.
It is also worth noting that SSA geo for each of the size fractions is significantly less than the corresponding SSA BET . The ratio of SSA BET /SSA geo , or roughness factor (Wolff-Boenisch et al., 2004), for the Grasåsen biotite ranges between $8 for the 25-53 lm size fraction and $3 NC, Not Calculated. Iron release appeared to be independent of grain size and geometry due to supersaturation of the fluid phase with respect to iron oxyhydroxides. for the 180-500 lm size fraction. The biotite surface consists of relatively smooth basal surfaces and rough edges (see Fig. S1), so when the grains are predominantly free of smaller particles it follows that the roughness factor is directly linked to SA edge . This also suggests that the values of SA edge may underestimate the actual surface area present on the edge surfaces of biotite grains. If the SA edge values were larger, then the calculated edge reactivity would be significantly lower. The use of initial specific surface area for normalising phyllosilicate dissolution rates in 12 h closed-system experiments appears to be appropriate as there is minimal change in the specific surface area between pre-and post-reaction samples. The use of the initial surface area in experiments of longer duration, either closed-or open-system may not be appropriate for phyllosilicate dissolution studies.

Stoichiometry of dissolution
The stoichiometry of element release from the biotite structure during the first 12 h of reaction in the closed-system experiments can be assessed by plotting the stoichiometry and surface area normalised moles of each element released from biotite (m 0 i ) against the corresponding moles of Si ( Fig. 6 and Fig. S6), and calculating the m 0 i /m 0 Si ratios (Table S5). In the organic ligand-free closed-system dissolution experiments the ratio of m 0 Al /m 0 Si (Fig. 6A) was pH independent and far from stoichiometric, revealing an average ratio of 2.2 (Standard Deviation, SD, 0.5). This non-stoichiometric release is also valid for the m 0 Fe /m 0 Si and m 0 Mg /m 0 Si (Fig. S6) ratios, which gave averages of 2.0 and 2.7, respectively (SD 0.8 and 0.6). It is worth noting that towards neutral pH the ratios of m 0 Al /m 0 Si and m 0 Fe /m 0 Si decrease (see insets of Fig. 6A and Fig. S6). At pH 6 the ratios for these elements are 1.04 and 0.48 respectively (SD 0.16 and 0.17), indicating near-stoichiometric release of Al and retention of Fe in the solid phase. This approach to stoichiometric Al release and Fe retention in the solid phase at pH 6 could be due to the precipitation of secondary Al and Fe oxy-hydroxide phases.
In the all open-system Grasåsen biotite experiments (Table S5), the ratio of m 0 i /m 0 Si , decreases during each experiment but in general remains greater than 1 after 300 h of reaction. Values reported in Table S5 are the average ratios during each experiment type. In one experiment at pH 3, the m 0 Al /m 0 Si ratio reached 0.7 after $250 h. Non-stoichiometric release of structural metals has previously been described for biotite (Acker and Bricker, 1992;Turpault and Trotignon, 1994;Kalinowski and Schweda, 1996;Malmströ m and Banwart 1997).

A Ligand Free
B Citric Acid C Oxalic Acid D DFOB The m 0 i /m 0 Si ratios from the closed-system experiments in the presence of organic ligands are presented in Fig. 6 and Fig. S6. In general, the m 0 i /m 0 Si ratios increase in the presence of organic ligands with respect to the ligand-free experiments (Table S5). With citric acid, oxalic acid, or DFOB present in the experimental fluid the ratios for m 0 Al /m 0 Si and m 0 Fe /m 0 Si at pH 5 and 6 are more consistent with the respective ratios at lower pH (Fig 6 and Fig. S6). This enhanced release of metals in the presence of organic ligands likely stems from the increased solubility of Fe and Al (Fig. S4). These plots show that the presence of organic ligands allow the non-stoichiometry of biotite dissolution to be consistent from pH 2.5 to 6. The shift to stoichiometric dissolution observed for Al and Fe in the ligand-free experiments at higher pH does not occur in the presence of the organic ligands due to the decrease in saturation indices of secondary Fe and Al phases in the fluids.

Dissolution mechanism
In Bray et al. (2014) we demonstrated the formation of Mg and Fe depleted tetrahedral and Al depleted octahedral layers at the biotite surface at low pH. The average depth of this depleted layer was calculated from (Bray et al., 2014): The average depth of removal of the ith element (Dep i ) was calculated by dividing the moles of the ith element removed from biotite during dissolution (m i ) by the average density of atoms in the biotite structure (atoms nm À3 ), where n i is the stoichiometric coefficient of the ith element in the bulk biotite, q biotite (g cm À3 ) represents the biotite density (3.09, average from webmineral.com), m biotite refers to the molecular mass of biotite (450.15 g mol À1 for the Grasåsen biotite), and N is Avogadro's constant (6.022 Â 10 23 moles À1 ). Based on these assumptions we calculated depletion depths from the total moles of elements released, ignoring the relative reactivities of the basal (0 0 1) and edge (h k 0) surfaces. Combining this information with the dissolution data from this study we can also calculate the depth of element removal from biotite from both the basal (00 1) and edge (hk0) planes. This can be done by replacing m 0 i in Eq. (8) with the released moles of elements from the basal and edge surfaces (m 0 edge,i and m 0 basal,i , Eq. 7). To visualise these removal fronts during biotite dissolution for each element, on the more reactive edge surfaces, at the various pH values and both in the presence and absence of organic ligands schematics showing dissolution fronts Fig. 7. Calculated depths of dissolution fronts in the tetrahedral and octahedral layers as a function of pH and organic ligand. Calculations based on the reactivity of (h k 0) and released concentrations after 12 h. linked to tetrahedral and octahedral layers were plotted in Fig. 7. We have shown (Table 5) that the edge surfaces are 120 times more reactive than the basal surfaces and thus the majority of dissolution occurs perpendicular to the (0 0 1) plane. The extent of element removal from biotite after 12 h as a function of pH and presence of organic ligands based on released element concentrations were attributed to either tetrahedral or octahedral sites depending on the biotite structure. At all pH in both organic ligand bearing and ligand-free experiments, octahedral metal (Fe, Mg) release was faster than tetrahedral element (Al and Si) release, consistent with numerous multi-oxide silicates (c.f. Oelkers et al., 2009). In the organic ligand-free experiments, Fe was preferentially released over Mg from the octahedral layer at pH 2.5-4, whereas at pH 5 and 6, Fe was preferentially retained over Mg. This change in element release is likely due to the control of secondary phase equilibria on Fe release above pH 5, where Fe-oxyhydroxide solubility is low. It is unclear from these experiments, therefore, if Fe retention was within the biotite structure or as a secondary mineral precipitate (in Fig. 7 it is represented as in the biotite structure) as in almost all experiments no quantitative microscopic evidence of secondary mineral formation is available. The preferential retention of Fe in the solid phase at near neutral conditions was observed in the presence of oxalic acid only at pH 6 ( Fig. 7), but not in the presence of citric acid or DFOB at any pH, probably due to their effect on increasing Fe solubility. It is also likely that during dissolution, Fe (II) oxidation within the biotite structure occurs. Such a process has been shown to lead to the formation of Fe (III) subdomains (Bonneville et al., 2009); however, the quantification of changes in the Fe oxidation states in the biotite samples was not quantified in this study. Similarly, the release of Al from tetrahedral sites in the presence of organic ligands was greatly increased compared to the ligand-free experiments above pH 3. The depth of Al removal at pH 6 in the presence of organic ligands was approximately 2.5 times greater than in the ligand-free experiments. Biotite dissolution is known to proceed non-stoichiometrically (Kalinowski and Schweda, 1996;Malmströ m et al., 1996;Malmströ m and Banwart, 1997;Voinot et al., 2013;Bray et al., 2014), however Acker and Bricker (1992) suggested that the dissolution of octahedral and tetrahedral layers can be internally stoichiometric, i.e., Fe and Mg release from the octahedral layer could be stoichiometric but overall this could be non stoichiometric compared to the bulk chemistry. Fig. 7 also demonstrates that for the Grasåsen biotite, this is not the case. Indeed, Mg and Fe release from the octahedral layer was not stoichiometric and similarly Al and Si release from the tetrahedral layer was also non stoichiometric.

CONCLUSION
The results summarised above illustrate the complex, but systematic dissolution behaviour of biotite, a common sheet silicate mineral. Dissolution is dominated by the biotite edges, which are between 45 and 120 times more reactive than the basal surface. Dissolution at these edges at acidic conditions occurs via the independent removal of the cations at distinct rates, where Mg and Fe are in general released faster than Al, which itself is released faster than Si. As a result, with time the biotite edges change composition and structure, the outermost edge consisting of an Al depleted tetrahedral framework adjacent to a Mg, Fe depleted octahedral layer. Deeper into the biotite, increasing Al, Mg, and Fe are present until a stoichiometric biotite is found deep within the grains. As rates are dominated by removal of material from edge surfaces, which constitute only a small fraction of the total biotite surface area, the degree to which total surface area normalised rates can represent natural processes is questionable.
Calculated dissolution rates from closed-system experiments converge with steady state dissolution rates derived from open-system experiments after approximately 12 h of reaction. The stoichiometry and surface area normalised dissolution rates of the Grasåsen and Moen biotites are largely consistent with existing data. However, the apparent rate constants calculated for biotite samples of differing chemical composition vary between elements, indicating element specific rates which could be described by element specific rate laws.
The presence of organic ligands, at low concentrations, enhanced the release of di-and tri-valent metals from both octahedral and tetrahedral layers, compared to ligand-free experiments. This was likely facilitated by the formation of metal-organic ligand complexes, driving the system equilibrium to promote further release of, in particular, Al and Fe. As such the relatively low concentrations of organic ligands considered in this study, similar to concentrations known to be exuded by microorganisms, increase the release rates of important secondary nutrients (Al, Fe, and Mg) from biotite. However, this increased release of metals did not significantly influence the biotite dissolution rate, based on Si release, as element reactivities and dissolution rates in biotite appear to be largely independent.